Effective questionnaire-based prediction models for type 2 diabetes across several ethnicities: a model development...

Area of Science:

• Diabetes research
• Epidemiology
• Public health

Background:

• Type 2 diabetes (T2D) disproportionately impacts non-White populations.
• Early detection and prediction are crucial for managing T2D.
• Large-scale, accessible tools are needed for risk stratification.

Purpose of the Study:

• To develop and validate questionnaire-based prediction models for T2D prevalence and incidence.
• To assess model performance across multiple ethnicities.
• To compare questionnaire-only models with existing clinical tools.

Main Methods:

• Logistic regression models trained on UK Biobank (White population).
• Validation in diverse ethnic groups within UK Biobank and externally in Lifelines cohort.
• Evaluation using Area Under the Receiver Operating Characteristic Curve (AUC).

Main Results:

• High accuracy for T2D prevalence (AUC=0.901) and incidence (AUC=0.873) in White population.
• Consistent performance across ethnicities (prevalence AUCs 0.855-0.894, incidence AUCs 0.819-0.883).
• Models outperformed existing non-laboratory risk tools, improving case reclassification.

Conclusions:

• Questionnaire-based models provide accurate, scalable prediction of T2D.
• These models can be implemented widely for population-wide risk stratification.
• Further validation in diverse global populations is warranted.